Examples of pixel binning technique and applications

Combining many pixels

CCD sensors can be read in different ways to improve the signal-to-noise ratio.

CCD sensors are widely used in astrophotography. Depending on the requirements, charges can be combined in different ways to improve the signal-to-noise ratio (SNR) and also speed up the readout process. A technique known as pixel binning is most often used.

Pixel binning is the combination of signals from multiple adjacent pixels. Typically, squares of 2×2 or 3×3 adjacent pixels are combined, which is known as 2x2 or 3x3 binning. The resolution of the image is reduced as a consequence. Whether this is beneficial or justified depends on the situation.

There are several sources of noise in a CCD image. Besides pure photon statistical noise, some noise is generated by the camera itself. In addition to thermal noise, it is generated by signal readout and amplification and affects each pixel individually. This is known as readout noise. Taking an area of 2×2 pixels, without CCD binning this noise affects all four pixels. If the values of the four pixels are combined retrospectively, the noise of the combined pixels can be defined as the square root of the sum of the noise squared:

σtotal=σ12+σ22+σ32+σ42+

The values σ1 to σ4 represent the noise electrons of the individual pixels, and σtotal is the noise of the entire unbinned area. Thus, one electron of noise per pixel becomes two electrons after they are combined. Assuming that each of the four pixels considered contained one electron object signal, this sums up linearly to four electrons. The SNR is therefore 2:4.

Binning changes the process flow. Before any enhancement is performed, the accumulated charges of the pixels in the register are added together. Thus, the noise occurs only once, in the previous example one electron of noise versus four electrons of signal. The SNR of the readout noise is therefore 1:4; better by a factor of 2.

Many image processing programs offer the option of retroactive binning. However, this software binning is missing an element of the desired effect. The SNR improves, but without optimising the readout noise.

CCD binning can be used to simplify and accelerate deep sky photography workflows. Thanks to the shorter readout and download times, rough focusing with 3×3 binning is very fast. The improved SNR means that even faint nebulae become visible with just a few seconds of exposure time, which makes it easier to adjust the frame. 2x2 binning is also often used in the creation of LRGB composites. It takes advantage of the fact that the RGB channels serve only to add colour to the image, and sharpness comes from the luminance. Lower resolution in the RGB channels is therefore not noticeable, which is why 2x2 binning is often used here in order to obtain a better SNR and to reduce colour noise.

There are other situations where binning makes sense, especially for astrometry: this includes, for example, the search for asteroids. This discipline has a special requirement because the objects change their position noticeably in a relatively short time. After just a few minutes, the object may have moved so far that it has changed pixel. An even longer exposure time does not bring any benefits. If the object is very faint, it may be invisible among the noise. If the task is to measure very faint small planets, binning is a possible solution, although the resulting reduction in resolution makes the position determination less accurate.

Binning is also very interesting for spectroscopy when the image of a slit is split into spectral colours. A high sampling rate orthogonal to the slit is relevant for high spectral resolution. In this way, pixels can be binned along the slit image to improve SNR without loss of resolution.

The graphic shows a schematic representation of the normal readout mode on the left and with 2x2 binning on the right. First, light from a star falls on 2×2 pixels and each collect ten electrons. Then the readout process begins by moving the electrons to the readout register (red). The differences begin in the third step. Without binning, the charges for each pixel are shifted right into the readout amplifier. With binning, the next row is immediately moved to the register and added to the first. By shifting them horizontally into the register, the binning process collects the charges from all four pixels in the readout amplifier. This completes the process, whereas the readout of the individual pixels takes two more steps and therefore takes longer.

True binning is not possible with colour sensors because adjacent pixels reflect different colour filters and cannot be combined in a meaningful way. Even if image processing programs offer this, it is a retrospective software binning.

Furthermore, the CMOS chip is generally not suitable for true binning. Due to its basic structure, all pixels are read out and amplified individually, so that the read-out noise for each pixel will always arise. The readout procedure required for binning does not exist here. Therefore, pixel binning on CMOS chips does not produce the desired effect of reduced readout noise. Here, CCD chips clearly have the advantage.

Author: Mario Weigand / Licence: Oculum Verlag GmbH